Recursive-STARK-based bandwidth-efficient mempool

If we were to adopt UTXOs as a natively-supported transaction-construct, couldn’t this be used to recursively aggregate users’ proofs of their UTXO spends, meaning we can then just include the 128kB root proof in a block and effectively support an enormous number of UTXO spends?

This works because UTXOs are stateless, which means we can just aggregate as many as possible on the memory layer and stream them into blocks as an aggregated proof.